Repository navigation
compiler modeling roadmap updates - #211
Merged
Merged
Conversation
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Add three new architectural sections to ROADMAP.md: - Compositional Basis: what is primitive (Node, Conj/Disj, cardinality, slots, recursion, collections), what the kernel honestly knows today (Int/String/Bool as leaves), structural constructors table, derived concepts table. - Composition Stack (Layers 0-7): full chain from Classical=True|False through Bit/Byte/Int/String/List up to Tree/DAG/Annotated (Layer 4, missing from std) to compiler domain types. Identifies Node as Tree<NodeFacts>, ModuleGraph as DAG<Module>, TypedModule as Annotated<Module, TypeFacts>. Type Layer Downgrade metric for architectural debt. P3.8 note: Tree<A> expressible today. - Rewrite "Phase 5 Parallel: Compositional Pipeline" around fact composition contracts. Tightened per review: "no compensating re-derivation" (not "no passes"), structurally authoritative (not fully-resolved), Optional binding/coproduct distinction preserved, SCC condensation for C4, C1 narrowed to truly timeless vocabulary, EmitGraphInfo carries rendering metadata not type authority. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Move 7 timeless structural functions from 04_types.dag to 00_core.dag: - Constructors: leaf_node, with_optional_cardinality, with_required_cardinality, no_span - Predicates: node_is_product, node_is_coproduct, node_has_structure Update imports across 14 .dag files to source these from v2.std.core. Bridge-era policy helpers (rt_type, error_type_node, child_inferred_or_name) stay in 04_types.dag — they encode stage-boundary semantics. Tighten Composition Stack per review: - Layer 1 uses Vector<n, Bit> not List<Bit> (machine/container distinction) - DAG<Id,A> defined as nodes + edges + acyclicity law, not "tree with sharing" - Node split: NodePayload (declared shape) vs NodeFacts (computed metadata) - ASTNode = Annotated<LabeledTree<NodePayload>, NodeFacts> - Pair/Either are ergonomic aliases over Conj/Disj, not new primitives - Span<I> is generic; SourceSpan = Span<Nat> - Token is parser-domain, not universal std - LabeledTree moved to Layer 4 (structural, not text/source) - Phase placement: declare Tree<A>, Annotated<A,F> in Phase 3 when generics stable Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
C2.1: Variant constructor locals now carry the structural parent enum node (binding.resolved / item) instead of leaf_node(name: binding.name). This means .inferred on variant expressions will be structurally authoritative. Two sites fixed: imported_variant_locals in 04_infer.dag and variant_locals_from_items in 04_items.dag. C4.1: Service expansion now exits early when converged (total service count unchanged between passes) instead of always running 5 passes. Most codebases converge in 1-2 passes. All 119 non-ignored tests pass. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
When resolve_node_bounded encounters an alias node (leaf with .inferred
but no structure), follow ONE level of .inferred to return the structural
target. This is O(1), not recursive — topo ordering guarantees the
target is already resolved.
Without this, downstream consumers see an alias_node wrapper (e.g.
{ name: "UserId", inferred: Resolved { node: leaf("String") } }) and
must re-resolve through TypeEnv — a layer downgrade from structural
to name-based authority.
All 112 non-ignored tests pass.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
C2.2: Remove explicit leaf_node(name: "Optional") overrides for Some/None variant locals. The general variant constructor mechanism (C2.1) already handles Optional — kernel_optional is a Disj node in env.bindings, so imported_variant_locals registers Some and None with the structural parent. The explicit overrides were undoing C2.1's fix. C2.3: In resolve_node_bounded, when encountering an alias node (leaf with .inferred but no structure), follow ONE level of .inferred to return the structural target. O(1), safe with topo ordering. This eliminates the alias_node wrapper that forced emit to re-resolve through TypeEnv. C2.4 (ResolvedFuncSig) is likely fixed by C2.3 — the return type flows through resolve_item_types → resolve_optional_node → resolve_node_bounded (now with alias following) before reaching DeclaredFuncSig. All 112 non-ignored tests pass. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
Add total timing output before each of the 4 early-return sites (frontend error, resolve error, normalize error, typecheck error). Previously t_total was initialized but only printed on the happy path. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
… notes Record implementation findings from this session: - C2.3 was the key fix (in resolve_node_bounded, not build_type_env) - C2.4 was already fixed by C2.3 (resolution chain flows through) - C2.2 was redundant with C2.1 (general mechanism handles Optional) - C4.1 implemented as early convergence (full topo-order deferred) Add detailed C2.5 implementation notes for next worker: 12 call sites in 05_emit_rust.dag, per-site migration pattern, risk assessment, is_already_optional simplification guidance. Mark all completed items: C1.1-C1.3, C2.1-C2.4, C4.1, C5.1. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
C1 moved node constructors from 04_types.dag to 00_core.dag. After stage0 regeneration, these functions live in v2_core instead of infer_types. Update test imports to match. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
|
|
||
| fn rust_map_value_type(receiver_type: Node, scope: InferScope) -> Node? { | ||
| let resolved = resolve_scrutinee_type_node(env: scope.type_env, n: normalize_access_type_node(n: receiver_type)) | ||
| let resolved = normalize_access_type_node(n: receiver_type) |
There was a problem hiding this comment.
Redundant double normalization in map value extraction
Low Severity
rust_map_value_type now calls normalize_access_type_node twice on the same input. Previously, the first line called resolve_scrutinee_type_node (which could produce a different node needing normalization). After replacing it with just normalize_access_type_node, the second call on resolved is redundant since normalize_access_type_node is idempotent — map_type will always equal resolved.
Additional Locations (1)
gunbai-bot Bot
pushed a commit
that referenced
this pull request
Oct 7, 2026
…bservation Repair 1 (msg_6a505028): megarac_recorder_trigger_power_on_field, megarac_trigger_power_on_field_wire, megarac_recorder_reset_survival and their rows and witness derive from the AMI bundle read; moved to gunbc-private #211, no public consumer remained. Repair 2: CdErrorDeviceEjected promoted a coexistence to a cause. The code is now CdErrorNonzero { wire, firmware }; stopped_with_reason, stop_reason_text and MegaRacPresentationStopped are deleted, and admitting the attach for 16 on 0.32 is an explicit policy (cd_error_attach_policy_admits). Co-Authored-By: Claude Sonnet 5.5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Note
Medium Risk
Touches compiler type resolution/inference and Rust emit by changing how aliases/variants are resolved and removing emit-time re-resolution; mistakes could surface as incorrect type-driven codegen or missed option/alias handling. Changes are localized and backed by new alias-chain regression coverage, but still affect core compilation behavior.
Overview
Strengthens the compiler’s fact composition so downstream stages rely on structurally resolved nodes instead of re-resolving by name.
Moves core
Nodestructural helpers (e.g.leaf_node, cardinality wrappers, product/coproduct predicates) into00_core.dagand updates imports accordingly. In inference, variant constructor locals now carry the parent enum’s structural node (and removes special-casing forOptional’sSome/None), andresolve_node_boundednow follows one level of alias.inferredto return the structural target.Simplifies emit by removing Rust emitter usage of
resolve_scrutinee_type_node/emit-time record-lit inference, pushing recursive-type metadata intoEmitGraphInfo(so emit scopes can use an emptyTypeEnv), and updatestype_needs_rcto operate withoutTypeEnv. Adds a regression test for chained type-alias field access and expands the roadmap documentation around compositional basis/pipeline.Written by Cursor Bugbot for commit 1aae3cb. This will update automatically on new commits. Configure here.